DDA-Net: Accurate TDD Channel Estimation via Deep Unfolding the Doppler-Delay-Angle Representation of Channel Signals

TL;DR

DDA-Net is a deep unfolding network that improves TDD massive MIMO channel estimation by explicitly modeling Doppler, delay, and angle sparsity, achieving significant accuracy gains with few training samples.

Contribution

It introduces a model-driven deep unfolding approach that combines physical data consistency with learned priors for efficient channel estimation under sparse pilots.

Findings

DDA-Net outperforms baselines with over 5 dB NMSE improvement at 10 dB SNR.

It maintains a 1.5 dB lead under zero-shot testing on 3GPP channels.

Ablation shows 3D window processing and Doppler parameterization are crucial for performance.

Abstract

In TDD massive MIMO systems, channel estimation under sparse frequency-hopping pilots is challenging: each snapshot captures only one narrow pilot block that hops across frequency, with tens of milliseconds between adjacent snapshots. Finite-window leakage and off-grid effects weaken the ideal Doppler-delay-angle (DDA) sparsity, limiting both classical sparse recovery and purely data-driven approaches lacking an explicit structured transform-domain model. We propose DDA-Net, a model-driven 3D deep unfolding network for joint multi-snapshot channel state reconstruction. DDA-Net unfolds an ADMM formulation with an exact closed-form data-consistency update that avoids tensor inversion, learns the prior via a lightweight Doppler-domain denoiser, and uses delay oversampling to reduce basis mismatch. On QuaDRiGa UMa-NLOS, DDA-Net improves NMSE over the best baseline by more than 5 dB at 10 dB SNR, and retains a lead of about 1.5 dB under zero-shot testing on 3GPP CDL-B channels at the same SNR. Ablation studies show that window-level 3D processing is necessary across scenarios, while Doppler parameterization adds in-distribution gains and recovers a clear lead under scenario shift after few-shot fine-tuning with only 20 target-domain samples. These results demonstrate that combining exact physical data consistency with a learned DDA-domain prior is an effective and sample-efficient approach to channel state acquisition under sparse frequency-hopping pilots.